File size: 10,857 Bytes
e5f4c09 0cc1b2b e5f4c09 0cc1b2b e5f4c09 0cc1b2b e5f4c09 0cc1b2b e5f4c09 0cc1b2b e5f4c09 0cc1b2b e5f4c09 0cc1b2b e5f4c09 0cc1b2b e5f4c09 0cc1b2b e5f4c09 0cc1b2b e5f4c09 642c04b e5f4c09 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 | """Canonical model, checkpoint-selection, and Hugging Face repository registry."""
from __future__ import annotations
from pathlib import Path
from typing import Any
NAMESPACE = "nishan-chatterjee"
COLLECTION_SLUG = (
"nishan-chatterjee/aspect-based-sentiment-analysis-6a9016a6d9cab7b093f122d3"
)
TOOLKIT_REPO = f"{NAMESPACE}/aspect-based-sentiment-analysis"
LANGUAGES = ("hbs", "slovenian")
MODES = ("masked", "unmasked")
MODEL_SPECS: dict[str, dict[str, Any]] = {
"xlmr": {
"display_name": "XLM-R",
"backend": "encoder",
"base_model": "FacebookAI/xlm-roberta-base",
"local_base_dir": "xlm-roberta-base",
"max_length": 512,
"hf_repo": f"{NAMESPACE}/aspectbench-xlmr",
},
"han-xlmr": {
"display_name": "HAN-XLM-R",
"backend": "han",
"base_model": "FacebookAI/xlm-roberta-base",
"local_base_dir": "xlm-roberta-base",
"max_length": 96,
"max_sentences": 128,
"hf_repo": f"{NAMESPACE}/aspectbench-han-xlmr",
},
"longformer": {
"display_name": "XLM-R Longformer",
"backend": "encoder",
"base_model": "markussagen/xlm-roberta-longformer-base-4096",
"local_base_dir": "markussagen_xlm-roberta-longformer-base-4096",
"max_length": 4096,
"global_attention": True,
"hf_repo": f"{NAMESPACE}/aspectbench-longformer",
},
"mdeberta-v3": {
"display_name": "mDeBERTa-v3",
"backend": "encoder",
"base_model": "microsoft/mdeberta-v3-base",
"local_base_dir": "microsoft_mdeberta-v3-base",
"max_length": 512,
"hf_repo": f"{NAMESPACE}/aspectbench-mdeberta-v3",
},
"mt5": {
"display_name": "mT5",
"backend": "mt5",
"base_model": "google/mt5-base",
"local_base_dir": "google_mt5-base",
"max_length": 512,
"max_target_length": 8,
"hf_repo": f"{NAMESPACE}/aspectbench-mt5",
},
"slavic-specific": {
"display_name": "BERTić / SloBERTa",
"backend": "encoder",
"base_model": {
"hbs": "classla/bcms-bertic",
"slovenian": "EMBEDDIA/sloberta",
},
"local_base_dir": {
"hbs": "classla_bcms-bertic",
"slovenian": "EMBEDDIA_sloberta",
},
"max_length": 512,
"hf_repo": f"{NAMESPACE}/aspectbench-slavic-specific",
},
"bge-m3-mlp": {
"display_name": "BGE-M3 dense + MLP",
"backend": "bge",
"base_model": "BAAI/bge-m3",
"local_base_dir": "embeddings/bge-m3",
"max_length": 8192,
"hf_repo": f"{NAMESPACE}/aspectbench-bge-m3-mlp",
},
}
def _entry(
source: str | None,
run: int | None,
validation_macro_f1: float | None,
*,
reason: str | None = None,
expected_source: str | None = None,
) -> dict[str, Any]:
return {
"source": source,
"run": run,
"validation_macro_f1": validation_macro_f1,
"available": source is not None,
"unavailable_reason": reason,
"expected_source": expected_source,
}
CHECKPOINTS: dict[tuple[str, str, str], dict[str, Any]] = {
("xlmr", "hbs", "masked"): _entry(
"additional-tasks/xlmr_truncated/masked/serbian/best_model_0.pt",
0,
0.9247234996073129,
),
("xlmr", "slovenian", "masked"): _entry(
"additional-tasks/xlmr_truncated/masked/slovenian/best_model_2.pt",
2,
0.8357596148205779,
),
("xlmr", "hbs", "unmasked"): _entry(
None,
1,
0.9261041473729551,
reason=(
"Validation metrics identify the best unmasked run, but its trained "
"checkpoint file is absent from the source tree. Retraining or archive "
"recovery is required."
),
expected_source="models/xlmr/no_summary/serbian/best_model_1.pt",
),
("xlmr", "slovenian", "unmasked"): _entry(
None,
2,
0.8286127750538895,
reason=(
"Validation metrics identify the best unmasked run, but its trained "
"checkpoint file is absent from the source tree. Retraining or archive "
"recovery is required."
),
expected_source="models/xlmr/no_summary/slovenian/best_model_2.pt",
),
("han-xlmr", "hbs", "masked"): _entry(
"results/global-context-modelling/simplified-dart-xlmr/serbian/best_model_2.pt",
2,
0.9137775530809181,
),
("han-xlmr", "slovenian", "masked"): _entry(
"results/global-context-modelling/simplified-dart-xlmr/slovenian/best_model_1.pt",
1,
0.8108233024053296,
),
("han-xlmr", "hbs", "unmasked"): _entry(
None,
2,
0.9106720097631231,
reason=(
"Validation metrics identify the best aspect-marker run, but its trained "
"checkpoint file is absent from the source tree. Retraining or archive "
"recovery is required."
),
expected_source=(
"models/global-context-modelling/with-aspect-markers/serbian/best_model_2.pt"
),
),
("han-xlmr", "slovenian", "unmasked"): _entry(
None,
2,
0.8127192586145947,
reason=(
"Validation metrics identify the best aspect-marker run, but its trained "
"checkpoint file is absent from the source tree. Retraining or archive "
"recovery is required."
),
expected_source=(
"models/global-context-modelling/with-aspect-markers/slovenian/best_model_2.pt"
),
),
("longformer", "hbs", "masked"): _entry(
"reviews/longformer/masked/serbian/best_model_0.pt", 0, 0.9235562204487023
),
("longformer", "slovenian", "masked"): _entry(
"reviews/longformer/masked/slovenian/best_model_1.pt", 1, 0.8349811751435684
),
("longformer", "hbs", "unmasked"): _entry(
"reviews/longformer/unmasked/serbian/best_model_0.pt", 0, 0.9237049934972151
),
("longformer", "slovenian", "unmasked"): _entry(
"reviews/longformer/unmasked/slovenian/best_model_1.pt", 1, 0.8250958292470165
),
("mdeberta-v3", "hbs", "masked"): _entry(
"reviews/mdeberta/masked/serbian/best_model_0.pt", 0, 0.9150760776163819
),
("mdeberta-v3", "slovenian", "masked"): _entry(
"reviews/mdeberta/masked/slovenian/best_model_1.pt", 1, 0.8242447008523016
),
("mdeberta-v3", "hbs", "unmasked"): _entry(
"reviews/mdeberta/unmasked/serbian/best_model_0.pt", 0, 0.9190600266461267
),
("mdeberta-v3", "slovenian", "unmasked"): _entry(
"reviews/mdeberta/unmasked/slovenian/best_model_1.pt", 1, 0.8223988954252537
),
("mt5", "hbs", "masked"): _entry(
"reviews/mt5/masked/serbian/best_model_0.pt", 0, 0.889948707182005
),
("mt5", "slovenian", "masked"): _entry(
"reviews/mt5/masked/slovenian/best_model_1.pt", 1, 0.8012255203557931
),
("mt5", "hbs", "unmasked"): _entry(
"reviews/mt5/unmasked/serbian/best_model_1.pt", 1, 0.8942548569348862
),
("mt5", "slovenian", "unmasked"): _entry(
"reviews/mt5/unmasked/slovenian/best_model_2.pt", 2, 0.8070351391836906
),
("slavic-specific", "hbs", "masked"): _entry(
"reviews/slavic_specific/masked/serbian/best_model_1.pt", 1, 0.9151353276491324
),
("slavic-specific", "slovenian", "masked"): _entry(
"reviews/slavic_specific/masked/slovenian/best_model_2.pt", 2, 0.8545257213567837
),
("slavic-specific", "hbs", "unmasked"): _entry(
"reviews/slavic_specific/unmasked/serbian/best_model_0.pt", 0, 0.9250658853407349
),
("slavic-specific", "slovenian", "unmasked"): _entry(
"reviews/slavic_specific/unmasked/slovenian/best_model_2.pt", 2, 0.8638678818708625
),
("bge-m3-mlp", "hbs", "masked"): _entry(
None,
2,
0.8840815245920188,
reason="Validation metrics exist, but the selected trained dense MLP head file is absent.",
expected_source="models/bge-m3_mlp/masked/serbian/best_model_2.pt",
),
("bge-m3-mlp", "slovenian", "masked"): _entry(
None,
1,
0.7514188081139693,
reason="Validation metrics exist, but the selected trained dense MLP head file is absent.",
expected_source="models/bge-m3_mlp/masked/slovenian/best_model_1.pt",
),
("bge-m3-mlp", "hbs", "unmasked"): _entry(
None,
1,
0.8894579185247774,
reason="Validation metrics exist, but the selected trained dense MLP head file is absent.",
expected_source="models/bge-m3_mlp/whole/serbian/best_model_1.pt",
),
("bge-m3-mlp", "slovenian", "unmasked"): _entry(
None,
1,
0.7393257242124599,
reason="Validation metrics exist, but the selected trained dense MLP head file is absent.",
expected_source="models/bge-m3_mlp/whole/slovenian/best_model_1.pt",
),
}
def _apply_local_release_metadata() -> None:
"""Expose newly retrained slots before their metadata reaches the remote."""
models_root = Path(__file__).resolve().parents[1] / "models"
for model_name in MODEL_SPECS:
path = models_root / model_name / "availability.json"
if not path.is_file():
continue
try:
import json
entries = json.loads(path.read_text(encoding="utf-8"))["entries"]
except (KeyError, OSError, TypeError, ValueError):
continue
for entry in entries:
key = (model_name, entry.get("language"), entry.get("mode"))
weight = models_root / model_name / str(entry.get("weight_path", "")).split(
f"{model_name}/", 1
)[-1]
if key not in CHECKPOINTS or not entry.get("available") or not weight.is_file():
continue
CHECKPOINTS[key].update(
available=True,
unavailable_reason=None,
validation_macro_f1=entry.get("validation_macro_f1"),
run=entry.get("selected_split", CHECKPOINTS[key].get("run")),
)
_apply_local_release_metadata()
def model_spec(model_name: str, language: str) -> dict[str, Any]:
spec = dict(MODEL_SPECS[model_name])
for key in ("base_model", "local_base_dir"):
if isinstance(spec[key], dict):
spec[key] = spec[key][language]
return spec
def weight_path(model_root: str | Path, model_name: str, language: str, mode: str) -> Path:
return Path(model_root) / model_name / language / f"{mode}.pt"
def all_slots():
for model_name in MODEL_SPECS:
for language in LANGUAGES:
for mode in MODES:
yield model_name, language, mode, CHECKPOINTS[(model_name, language, mode)]
|